I have noticed that text cleaning is the most difficult part in OCR pipeline. I have struggled alot on this part, without properly cleaned text OCR simply fails in terms of accuracy. In order to handle text cleaning seperately I created a GitHub repo that uses AI to clean up all text in a image. Once the text is cleaned we can choose our own custom OCR models on it. I have personally seen OCR accuracy shoot up to 99% on a properly preprocessed and cleaned image.
Here is a Github: https://github.com/ajinkya933/ClearText link. Regards Ajinkya -- You received this message because you are subscribed to the Google Groups "tesseract-ocr" group. To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-ocr+unsubscr...@googlegroups.com. To view this discussion visit https://groups.google.com/d/msgid/tesseract-ocr/CAHy6iNOjhs7ZY7r26fGzqJOUr2e%2BF3bY%3DeDCHjM-VD7XH5M%3DTA%40mail.gmail.com.
